Your stack already has agents, catalogs, pipeline monitors, and metric definitions. Nodal doesn't replace any of them — it sits above them and answers the one question none of them do: is the number right, and will it stay right?
Bring the agent you already picked — or swap it next quarter. Nodal feeds it one governed context over MCP and tests its answers against your ground truth, so a model change is a benchmark run, not a leap of faith. Agent-agnostic means no lock-in: the truth stays yours even when the agent changes. Watch Claude Code and Codex land on the same governed answer →
A catalog describes what data exists. Nodal reads from your catalog as one context source — then measures whether answers built on it actually come back right, and how much each piece of documentation earns in answer quality. Complementary, not competing.
Pipeline observability catches freshness and volume anomalies — the loud failures. Silent SQL is the quiet one: the pipeline is green and the answer is wrong. Nodal regression-tests questions against verified results, so drift in meaning gets caught the way drift in data does.
A metric definition only helps if the agent actually uses it. Nodal sits above the semantic layer: it verifies that answers respect your definitions, re-runs the suite when a definition changes, and pins any drift to the commit that caused it.
Observability on every question, regression evals on every change, one governed context for every agent. How it works →